# DeepJ: Style-Specific Music Generation

**Authors:** Huanru Henry Mao, Taylor Shin, Garrison W. Cottrell  
**arXiv:** [1801.00887](https://arxiv.org/abs/1801.00887)  
**Date:** 2018-01-03  
**Summary page:** https://www.summarizepaper.com/en/paper/1801.00887v1/

## Key points

- Recent advances in deep neural networks have revolutionized music composition, enabling algorithms to generate music comparable to compositions by human musicians.
- Existing algorithms lack flexibility in adjusting specific parameters during the music generation process, limiting user customization.
- DeepJ is an innovative end-to-end generative model introduced by researchers Huanru Henry Mao, Taylor Shin, and Garrison W. Cottrell to address the need for customizable music generation tools.
- Key contributions of DeepJ include novel methods for learning musical style and dynamics within deep neural networks.
- DeepJ can generate music conditioned on a unique mixture of composer styles, allowing users to specify desired artistic influences for the generated piece.
- Incorporating style-specific parameters into the model architecture gives users unprecedented control over the generated music compared to traditional algorithms.
- The effectiveness of DeepJ was demonstrated through a proof-of-concept showcasing a technique for manipulating the style of generated music and outperforming existing approaches like Biaxial LSTM method according to human raters.
- Overall, DeepJ represents a significant advancement in AI-driven music composition by providing artists with a versatile tool for generating high-quality music tailored to specific stylistic preferences.

## Summary

Recent advances in deep neural networks have revolutionized the field of music composition, enabling algorithms to generate music that rivals compositions created by human musicians. However, a key limitation of existing algorithms is the lack of flexibility in allowing users to adjust specific parameters during the music generation process. This ability to fine-tune properties of generated music is crucial for assisting artists, filmmakers, and composers in their creative endeavors. In response to this need for more customizable music generation tools, a team of researchers including Huanru Henry Mao, Taylor Shin, and Garrison W. Cottrell introduced DeepJ - an innovative end-to-end generative model designed to compose music based on a specific blend of composer styles. The key contributions of their work include novel methods for learning musical style and dynamics within the context of deep neural networks. One notable feature of DeepJ is its capability to generate music conditioned on a unique mixture of composer styles, allowing users to specify the desired artistic influences for the generated piece. By incorporating these style-specific parameters into the model architecture, DeepJ offers a level of control over the generated music that was previously unattainable with traditional algorithms. The researchers demonstrated the effectiveness of DeepJ by showcasing a simple yet powerful technique for manipulating the style of generated music as a proof-of-concept. Through evaluation with human raters, they were able to show significant improvements over existing approaches such as the Biaxial LSTM method. Overall, DeepJ represents a significant advancement in the field of AI-driven music composition by providing artists and creators with a versatile tool for generating high-quality music tailored to specific stylistic preferences. As machine learning continues to evolve, innovations like DeepJ hold great promise for enhancing creativity and collaboration across various artistic disciplines.